I had both a very similar and very different experience with my father.
To me the main difference is the role of the other parent. My mother was always very protective of us, and as far as she could would sympathise and not try and make us see a person who clearly wasn’t there, or mythologise about a parent who could have been. My older siblings forgave my father (and had memories of him when he was more normal), but I never did, and, I think most importantly, nobody asks me to. Fuck that guy.
I guess what I’m trying to say is that this may be much more about the other parent. The author’s father is in deep denial and isn’t trying to protect his kids, and is the core of this problem. Let the kids grieve the mother they never had, rather than endless apologism.
I use it purely for finding stuff in text documents and OCR, at which it excels, but I've found that for any kind of analysis it's some bewildering mixture of smug, defensive, paranoid, cheerful, and wrong about everything.
I feel like the headline here is "the Artificial Analysis index is becoming an increasingly poor measure of capability, having previously been quite good".
One recent project that LLMs has made easy has been properly segmenting my home network, which my boss archly described as "the internet of everything" for the sheer number of weird devices on it including a Brother printer under my bed.
I have a special sandboxed agent with encrypted credentials (and on its own locked down physical box) that's set me up vLANs and the like on a bunch of MikroTic hardware that I wouldn't have bothered to do myself if I'd had to be hand-maintaining it.
Been connecting a Raspberry Pi Zero to various points of it via ethernet and wireless to scan the intended segmentation works fine, and it does look that way.
> Maintainers get their FOSS projects bombed with AI slop PRs and issues, and you call their position to not accept slop a tribalist
Sure, and that's the motte.
This is a direct quote from the linked website:
> Any software whose code, documentation, binaries, or other assets include, in any part, the output of a generative AI tool such as an LLM, is considered “tainted” for our purposes.
As someone with a highly uneven and kinda basic maths background, I'd have thought that simplification of a proof, and pulling things into a coherent whole would be exactly what computers excel at, but everything I've seen (or heard) suggests that's not really the case. Are you able to explain why?
There are some (typically small) simplifications that computers are indeed good at. These could be thought of as analogous to compiler optimizations - they’ll see that your proof is basically composing several lemmas and that some of the work in the middle can be removed so it can just do that. In many cases with lean if you post a short to medium-length proof into an llm it can suggest these types of small improvements for you.
What they’re not that great at is the type of improvements that really make a big difference: big simplifications etc. This is not super suprising when you understand how this works.
Firstly proofs in lean (and I think many other assistants) are essentially functions which take hypotheses and instantiate a particular type. That type being equivalent (due to the Curry-Howard correspondence) to a proof of some proposition. The assistant uses dependent type theory to make sure all the types match up at each step, which is what makes it rigorous. So the task of simplifying a proof is basically finding a route through an infinite-dimensional space where you know your starting location and the desired end point, but at any step you can go in basically any direction you like so long as you can justify your step. You justify a step by either choosing from a (vast at this point) library of lemmas and theorems that are in your standard library, or by recursively building the proof/function you need to justify the step you want to take using this same process you’re trying to use for simplification. This is incredibly flexible, but means you’re actually spoiled for choice and planning is hard.
Secondly, in “pen and paper” proofs, LLMs don’t know the difference between steps which are important/easy/hard/intuitive/non-intuitive for human mathematicians, so lots of the proofs they produce are extremely clunky and pedantic in places and dwell for ages over details that are obvious and would usually be skipped over and then will move very fast through things that justify more attention. This unevenness is part of what makes their proofs hard for humans to work through, but also means they are generally poor judges of what would actually constitute a decent simplification.
Uhm three ideas about that (they don't solve the problem on their own, they just show mathematicians care about this issue)
Category theory can be viewed as a framework to make math more modular and compatible. For example topos theory can be used to transport definitions to other "mathematical universes" of sorts
Another point of view is that is working with synthetic mathematics which is a bit like "programming" mathematics with an abstract API. The opposite, analytic mathematics, leak implementation details (for example, the von Neumann construction of natural numbers, which is analytic, leak the implementation detail that 0 is defined as the empty set; but the peano axioms are synthetic and doesn't leak any such details)
There is also the principle that two mathematical objects that are isomorphic (have the same structure) are actually the same thing. For example if you construct the natural numbers two different ways, they are "the same thing" in a sense. This usually is stated informally, but a rigorous formulation of that is the univalent principle
Some books say in the first page that they will assume all groups are Abelian/Commutative, and not repeat the assumption in each page.
There are like 8 conventions to define the Fourier Transform. All are equivalent, but each one add different constants here and there. Beware to combine formulas from different books.
I guess it's easier for the computer to track the details and remember to translate the names in each book article to a consistent set of definitions.
> Your counterparty will have its own agent, a larger budget for that agent, and a lot more skin in the game to ensure they retain their advantage
I did some work on negotiating agents. This doesn’t have to be a zero sum game. If you can craft a per-user package effectively (for many services) then both sides end up with a better deal, and competition in the market stops the provider entity from dictating terms.
A lot of systems are either adversarial or the v0 agents are relying on a sort of free rider / human good will that won't scale when 100M people have agent bots.
Apples iPhone call answering for unknown numbers was OK until all the overseas robocallers made their robot respond somewhat conversationally and hang on the line so that it vaguely looks like a real enough call. I can still tell its a robot since I am not senile yet, but I still have all the notifications about the call/transcription/voicemail/etc to tap.
People deploying agents to call restaurants for reservations works for like the first 10 calls that restaurant gets and now they are going to hang up on everyone's bots. Many websites already blocking agents. Some websites where agents were scraping for reservations/inventory are outright locking peoples accounts. Etc.
This is neither a particularly contemporary problem nor an American one.
reply